U.S. patent number 7,302,087 [Application Number 10/656,885] was granted by the patent office on 2007-11-27 for daubechies wavelet transform of iris image data for use with iris recognition system.
This patent grant is currently assigned to Senga Advisors, LLC. Invention is credited to Seong-Won Cho.
United States Patent |
7,302,087 |
Cho |
November 27, 2007 |
Daubechies wavelet transform of iris image data for use with iris
recognition system
Abstract
The Present invention relates to a method of recognizing the
human iris corresponding to a field of a biometric technology, and
more particularly to a method of recognizing human iris using
Daubechies wavelet transform, wherein the dimensions of
characteristic vectors are reduced by extracting iris features from
inputted iris image signals through the Daubechies wavelet
transform, binary characteristic vectors are generated by applying
quantiztion functions to the extracted characteristic values so
that utility of human iris recognition can be improved since
storage capacity arid processing time thereof can be improved since
storage capacity characteristic vectors, and a measurement process
suitable for the low capacity characteristic vectors is employed
when measuring vectors and previously registered characteristic
vectors.
Inventors: |
Cho; Seong-Won (Seoul,
KR) |
Assignee: |
Senga Advisors, LLC (Boston,
MA)
|
Family
ID: |
19706518 |
Appl.
No.: |
10/656,885 |
Filed: |
September 5, 2003 |
Prior Publication Data
|
|
|
|
Document
Identifier |
Publication Date |
|
US 20040114781 A1 |
Jun 17, 2004 |
|
Related U.S. Patent Documents
|
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
Issue Date |
|
|
PCT/KR01/01303 |
Jul 31, 2001 |
|
|
|
|
Foreign Application Priority Data
|
|
|
|
|
Mar 6, 2001 [KR] |
|
|
2001-11440 |
|
Current U.S.
Class: |
382/118;
382/117 |
Current CPC
Class: |
G06K
9/00597 (20130101) |
Current International
Class: |
G06K
9/00 (20060101) |
Field of
Search: |
;382/117,118 |
References Cited
[Referenced By]
U.S. Patent Documents
Foreign Patent Documents
Other References
Daubechies, Ingrid; Antonini, Marc; Barlaud, Michel; Mathieu,
Pierre; Image Coding Using Wavelet Transform; Apr. 1992; IEEE; vol.
1, No. 2; p. 205-220. cited by examiner .
Boles, et al., "A Human Identification Technique Using Images of
the Iris and Wavelet Transform", IEEE transactions on Signal
Processing, vol. 46, No. 4, pp. 1185-1188, Apr. 1998. cited by
other .
Boles, "A Security System Based on Human Iris Identification Using
Wavelet Transform", 1997 First International Conference on
Knowledge Based Intelligent Electronic Systems, pp. 533-541, May
21-23, 1997. cited by other .
Boles, "A Wavelet Transform Based Technique for the Recognition of
the Human Iris", ISSPA (International Symposium on Signal
Processing and its applications) Gold Coast, Australia, pp.
601-604, Aug. 25-30, 1996. cited by other.
|
Primary Examiner: Mehta; Bhavesh M
Assistant Examiner: Schaffer; Jonathan
Attorney, Agent or Firm: Knobbe Martens Olson & Bear,
LLP
Parent Case Text
RELATED APPLICATIONS
This application is a continuing application under 35 U.S.C. .sctn.
365 (c) claiming the benefit of the filing date of PCT Application
No. PCT/KR01/01303 designating the United States, filed Jul. 31,
2001. The PCT Application was published in English as WO 02/071317
A1 on Sep. 12, 2002, and claims the benefit of the earlier filing
date of Korean Patent Application No. 2001/11440, filed Mar. 6,
2001. The contents of the Korean Patent Application No. 2001/11440
and the international application No. PCT/KR01/01303 including the
publication WO 02/071317 A1 are incorporated herein by reference in
their entirety.
Claims
What is claimed is:
1. A method of processing iris image data, comprising: providing
data representing an image of an iris of an eye; performing a
Daubechies wavelet transform on the iris image data so as to create
multiple pieces of transformed image data; selecting one from the
multiple pieces of transformed image data; repeating to perform the
Daubechies wavelet transform on a piece of image data so as to
create additional multiple pieces of transformed image data and
subsequently to select one from the additional multiple piece of
the transformed image data, wherein the piece of image data on
which the Daubechies wavelet transform is repeated is the selected
piece that is selected in the immediately previous selecting step,
wherein performing the Daubechies wavelet transform and
subsequently selecting are repeated at least once; upon completion
of the repeating step, forming a characteristic vector of the iris
image of the eye, wherein the characteristic vector comprises
quantatized pixel values of the selected piece of transformed image
data that is selected in the last selecting of the repeating step,
wherein the quantitized pixel values comprise at least two positive
values and at least two negative values; providing a reference
characteristic vector of a registered iris image; computing an
inner product of the reference characteristic vector and the
characteristic vector of the iris image of the eye; determining
whether the iris image of the eye matches with the registered iris
image based on a value of the inner product.
2. The method of claim 1, wherein the quantitized pixel values
comprise one of the at least two positive values has the same
absolute value as one of the at least two negative values.
3. The method of claim 1, wherein the quantitized pixel values
comprise a first positive value and a second positive value,
wherein the second positive value is greater than two times of the
first positive value.
4. The method of claim 1, wherein each piece of transformed image
data comprise a lesser amount of data than the image data prior to
performing the Daubechies wavelet transform.
5. The method of claim 1, wherein the iris image is determined to
match the registered iris image when the inner product is greater
than a predetermined threshold value.
6. The method of claim 1, wherein the selected piece of transformed
image data which is selected in each selection step represents more
information on iris patterns than the other pieces of transformed
image data created in the Daubechies wavelet transform which is
performed immediately prior to each selection step.
7. The method of claim 5, the selected piece of transformed image
data which is selected in each selection step comprises more low
frequency components than the other pieces of transformed image
data created in the Daubechies wavelet transform which is performed
immediately prior to each selection step.
8. The method of claim 1, wherein each of the multiple pieces of
transformed image data created in each Daubechies wavelet transform
is classified based on frequency components of the data.
9. The method of claim 1, wherein each of the multiple pieces of
transformed image data created in each Daubechies wavelet transform
is classified based on frequency components of each piece of the
transformed image data in two perpendicular directions.
10. The method of claim 1, wherein each of the multiple pieces of
transformed image data created in each Daubechies wavelet transform
is classified one of HH, HL, LH and LL, wherein HH represents high
frequency components in a first direction and a second direction,
the first and second directions being perpendicular to each other,
wherein HL represents a high frequency component in the first
direction and a low frequency component in the second, direction,
wherein LH represents a low frequency component in the first
direction and a high frequency component in the second direction,
and wherein LL represents low frequency components in the first and
second directions.
11. The method of claim 9, wherein the characteristic vector
comprises information of at least one piece of the transformed
image data which are classified as HH.
12. The method of claim 10, wherein the information of the HH
comprises an average value of the piece of transformed image data
classified as HH.
13. The method of claim 9, wherein the characteristic vector
comprises information of the selected piece that is selected in the
lasts selecting of the repeating step is classified as LL.
14. The method of claim 13, wherein the information of the LL
comprises a substantial portion of the transformed image data of
the last selected data piece.
15. The method of claim 12, wherein the information of the LL
comprises all of the transformed image data of the last selected
data piece.
16. The method of claim 12, wherein a total number of the
Daubechies wavelet transform is N, wherein the characteristic
vector comprises an N-1 number of values of the HH data pieces.
17. The method of claim 1, wherein the number of repetitions is set
such that the total number of the Daubechies wavelet transform is
from 2 to 7.
18. The method of claim 1, wherein the number of repetitions is set
such that the total number of the Daubechies wavelet transform is
from 4.
19. A device for use in processing iris image data, comprising:
means for providing data representing an image of an iris of an
eye; means for performing a Daubechies wavelet transform on the
iris image data, thereby creating multiple pieces of transformed
image data, wherein the means for performing the transform is
configured to repeat the Daubechies wavelet transform on one of the
multiple pieces of transformed image data created in the
immediately previous transform; means for forming a characteristic
vector of one piece of transformed image data; a database
comprising a reference characteristic vector of a registered iris
image; means for computing an inner product of the reference
characteristic vector and the characteristic vector of the iris
image of the eye from the means for forming; and means for
determining whether the iris image of the eye matches with the
registered iris image based on the inner product.
20. A security system comprising: an input device configured to
receive data representing an image of an iris of an eye; a first
circuit configured to perform a Daubechies wavelet transform on the
iris image data and further configured to repeat the Daubechies
wavelet transform on a subset of the transformed iris image data a
predetermined number of times so as to generate a further subset of
transformed image data; a second circuit configured to form a
characteristic vector of the iris image comprises information of
the further subset of transformed image data from the first
circuit; a memory configured to store a reference characteristic
vector of a registered iris image; a third circuit configured to
compute an inner product of the reference characteristic vector and
the characteristic vector from the second circuit so as to
determine whether the iris image matches a pre-registered iris
image.
Description
FIELD OF INVENTION
The present invention relates to a method of recognizing the human
iris using Daubechies wavelet transform. More particularly, the
present invention is directed to a method of recognizing human iris
using Daubechies wavelet transform, wherein the dimensions of
characteristic vectors are reduced by extracting iris features from
inputted iris image signals through the Daubechies wavelet
transform, binary characteristic vectors are generated by applying
quantiztion functions to the extracted characteristic values so
that utility of human iris recognition can be improved since
storage capacity and processing time thereof can be reduced by
generating low capacity characteristic vectors, and a measurement
process suitable for the low capacity characteristic vectors is
employed when measuring the similarity between the generated
characteristic vectors and previously registered characteristic
vectors.
BACKGROUND OF INVENTION
An iris recognition system is an apparatus for performing
identification of each individual by differentiating iris patterns
of the pupil of an eye, which are unique for each individual. It
has superior identification accuracy and excellent security as
compared with other biometric method using voice and fingerprints
from each individual. A human iris is the region between a pupil
and a white sclera of an eye, and iris recognition is a technique
for performing identification of each individual based on
information that is obtained from an analysis of the iris patterns
which are different in each individual.
In general, it is a core technology to efficiently acquire unique
characteristic information from input images in the field of an
applied technology for performing identification of each individual
by utilizing the characteristic information of the human body. A
wavelet transform is used to extract characteristics of the iris
images, and it is a kind of technique of analyzing signals in
multiresolution mode. The wavelet transform is a mathematical
theory of formulating a model for signals, systems and a series of
processes by using specifically selected signals. These signals are
referred to as little waves or wavelets. Recently, the wavelet
transform is widely employed in the field of signal and image
processing since it has a fast rate as compared with a conventional
signal processing algorithm based on the Fourier transform, and it
can efficiently accomplish signal localization in time and
frequency domains.
On the other hand, the images, which are obtained by extracting
only iris patterns from the iris images acquired by image
acquisition equipment and normalizing the patterns at a
450.times.60 size, are used to extract characteristic values
through the wavelet transform. Further, a Harr wavelet transform
has been widely used in conventional iris recognition, image
processing and the like. However, Harr wavelet functions have
disadvantages in that the characteristic values are discontinuously
and rapidly changed and that high resolution of the images cannot
be obtained in a case where the images are again decompressed after
they have been compressed. On the contrary, since Daubechies
wavelet functions are continuous functions, the disadvantages of
the Harr wavelet functions that the values thereof are
discontinuously and rapidly changed can be avoided, and the
characteristic values can be extracted more accurately and
delicately. Therefore, in a case where the images are to be again
decompressed after they have been compressed by using the
Daubechies wavelet transform, the images can be restored in high
resolution nearer to the original images than when the Harr wavelet
transform is used. Since the Daubechies wavelet functions are more
complicated than the Harr wavelet functions, there is a
disadvantage in that more arithmetic quantity may be needed.
However, it can be easily overcome by the recent advent of
ultrahigh speed microprocessors.
There is also an advantage in that the Daubechies wavelet transform
can obtain fine characteristic values in the process of performing
the wavelet transform for extracting the characteristic values.
That is, if the Daubechies wavelet transform is used, expression of
the iris features can be made in a low capacity of data and
extraction of the features can be made accurately.
Methods of extracting the characteristic values and forming the
characteristic vectors by using Gabor transform been mainly used in
the conventional iris recognition field. However, the
characteristic vectors generated by these methods are formed to
have 256 or more dimensions, and they have at least 256 bytes even
though it is assumed that one byte is assigned to one dimension.
Thus, there is a problem in that practicability and efficiency can
be reduced when it is used in the field where low capacity
information is needed. Accordingly, it is necessary to develop a
method of forming the low capacity characteristic vectors wherein
processing, storage, transfer, search, and the like of the pattern
information can be efficiently made. In addition, since a simple
method of measuring a distance such as a Hamming distance (HD)
between two characteristic vectors (characteristic vectors relevant
to the input pattern and stored reference characteristic vectors)
is used for pattern classification in a prior art such as U.S. Pat.
No. 5,291,560, there are disadvantages in that formation of the
reference characteristic vectors through generalization of the
pattern information cannot be easily made and information
characteristics of each dimension of the characteristic vectors
cannot be properly reflected.
That is, in the method of using the Hamming distance in order to
verify the two characteristic vectors generated in the form of
binary vectors, bit values assigned according to respective
dimensions are compared with each other. If they are identical to
each other, 0 is given; and if they are different from each other,
1 is given. Then, a value divided by the total number of the
dimensions is obtained as a final result. The method is simple and
useful in discriminating a degree of similarity between the
characteristic vectors consisted of binary codes. When the Hamming
distance is used, the comparison result of all the bits becomes 0
if identical data are compared with each other. Thus, the result
approaching to 0 implies that the data belong to the persons
themselves. If the data do indeed belong to the person, the
probability of a degree of similarity will be 0.5. Thus, upon
comparison with the other person's data, it is understood that the
values converge around 0.5. Accordingly, a proper limit set between
0 and 0.5 will be a boundary for differentiating the data of the
persons themselves from the other person's data. The Hamming
distance (HD) is excellent in performance thereof in a case where
the information is obtained from the extracted iris features by
subdividing the data, but it is not suitable when low capacity data
is to be used. In other words, in a case where total number of the
bits of the characteristic vectors having 256-byte information is
2048, considerably high acceptance rates can be achieved even
though the Hamming distance is applied. However, in a case where
low capacity characteristic vectors in which the number of
characteristic vectors is reduced are used as in the present
invention, high acceptance rates cannot be obtained.
On the other hand, in a case where the low capacity characteristic
vectors are used, improvement of the acceptance rate is limited to
a certain extent since lost information is increased. Thus, a
method of preventing loss of the information while maintaining
minimum capacity of the characteristic vectors should be considered
in the process of generating the characteristic vectors.
SUMMARY OF THE INVENTION
One aspect of the present invention is to provides a method of
processing an iris image data. The method comprises: providing data
representing an image of an iris of an eye; performing a Daubechies
wavelet transform on the iris image data, thereby dividing the iris
image data into multiple data segments; repeating the Daubechies
wavelet transform a predetermined number of times on one of the
data segments divided in the immediately previous transform,
thereby dividing the data segment on which the transform is
performed into smaller data segments, wherein the data segment on
which the transform is performed represents more information on
iris patterns than the other data segments divided in the
immediately previous transform; and forming a characteristic vector
of the iris image comprising information of at least one data
segment divided in each Daubechies wavelet transform. In the
method, the data segment representing more information on iris
pattern than the other segments divided in the immediately previous
transform comprise more low frequency components than the other
segments. Each of the data segments produced in each Daubechies
wavelet transform is classified based on frequency components of
the data. Each of the data segments produced in each Daubechies
wavelet transform is classified based on frequency components of
the data in two perpendicular directions on an image each data
represents.
In the above-described method, the data segments produced in each
Daubechies wavelet transform is classified one of HH, HL, LH and
LL, wherein HH represents high frequency components in a first
direction and a second direction on an image each data represents,
the first and second directions being perpendicular to each other,
wherein HL represents a high frequency component in the first
direction and a low frequency component in the second direction,
wherein LH represents a low frequency component in the first
direction and a high frequency component in the second direction,
and wherein LL represents low frequency components in the first and
second directions. The characteristic vector comprises information
of a data segment characterized as HH divided in each of the
Daubechies wavelet transform. The information of the HH data
segment comprises an average value of data of the segment
representing the image thereof. The characteristic vector comprises
information of a data segment characterized as LL divided in the
last Daubechies wavelet transform. The information of the LL data
segment comprises a substantial portion of the data of the segment
representing the image thereof. The information of the LL data
segment comprises all of the data of the segment representing the
image thereof. A total number of the Daubechies wavelet transform
is N, the characteristic vector comprises an N-1 number of values
of HH data segments. The method further comprises quantitizing
values of the characteristic vector. The predetermined number of
repetitions is set such that a total number of the Daubechies
wavelet transform is from 2 to 7. The predetermined number of
repetitions is set such that a total number of the Daubechies
wavelet transform is from 4. The method further comprises
registering the characteristic vector with or without further
processing.
Another aspect of the present invention provides a device for use
in processing iris image data. The method comprises: means for
providing data representing an image of an iris of an eye; means
for performing a Daubechies wavelet transform on the iris image
data, thereby dividing the iris image data into multiple data
segments, wherein the means for performing the transform is
configured to repeat the Daubechies wavelet transform a
predetermined number of times on one of the data segments divided
in the immediately previous transform, thereby dividing the data
segment on which the transform is performed into smaller data
segments, and wherein the data segment on which the transform is
performed represents more information on iris patterns than the
other data segments divided in the immediately previous transform;
and means for forming a characteristic vector of the iris image
comprising information of at least one data segment divided in each
Daubechies wavelet transform. The method further comprises
processing the characteristic vector to determine whether the iris
image matches a pre-registered iris image. The characteristic
vector is processed together with a characteristic vector of the
pre-registered iris image to produce an inner product of the
characteristic vectors. The iris image is determined to match the
pre-registered iris image when an inner product of the
characteristic vector and a characteristic vector of the
pre-registered iris image is greater than a predetermined threshold
value.
Another aspect of the present invention provides a device for
processing iris image data. The device comprises: an input device
configured to receive data representing an image of an iris of an
eye; a first circuit configured to perform a Daubechies wavelet
transform on the iris image data a predetermined number of times;
and a second circuit configured to form a characteristic vector of
the iris image based on the Daubechies wavelet transform.
Still another aspect of the present invention provides a device for
identifying an iris pattern. The method comprises: means for
obtaining a characteristic vector from an iris image in accordance
with the above-described method of processing an iris image data;
and means for processing the characteristic vector to determine
whether the iris image matches a pre-registered iris image.
Still another aspect of the present invention provides a security
system. The system comprises: the above-described iris image data
processing device; and a third circuit configured to process the
characteristic vector to determine whether the iris image matches a
pre-registered iris image.
SUMMARY OF INVENTION
Therefore, the present invention is contemplated to solve these
problems mentioned above. An object of the present invention is to
provide a method of forming low capacity characteristic vectors
wherein a false acceptance rate FAR) and a false rejection rate
(FRR) can be remarkably reduced, as compared with a conventional
Harr wavelet transform, by extracting iris features from inputted
iris image signals through a Daubechies wavelet transform.
Another object of the present invention is to provide a method of
measuring the similarity between characteristic vectors, wherein
loss of information produced when low capacity characteristic
vectors are formed can be minimized and the low capacity
characteristic vectors can be properly used for the similarity
measurement.
In order to achieve the above objects of the present invention,
there is provided a method of recognizing the human iris using the
Daubechies wavelet transform, wherein preprocessing for extracting
only an iris image from an eye image of a user acquired by image
acquisition equipment using a halogen lamp illuminator is performed
and identification of the user is determined by the preprocessed
iris image comprising (a) repeatedly performing the Daubechies
wavelet transform of the preprocessed iris image at predetermined
times so as to multi-divide the iris image, and extracting an image
including high frequency components from the multi-dvided image so
as to extract iris features; (b) extracting characteristic values
of a characteristic vector from the extracted image including the
high frequency components, and generating a binary characteristic
vector by quantizing relevant characteristic values; and (c)
determining the user as an enrollee by measuring similarity between
the generated characteristic vector and a previously registered
characteristic vector.
The present invention will be briefly described below. The iris
image is acquired through the image acquisition equipment in which
the halogen lamp illuminator is used. By repeatedly performing the
Daubechies wavelet transform of the inputted iris image, the iris
image is multi-divided and iris features having optimized sizes
thereof are extracted. The characteristic vector, which is
effective in displaying and processing the image, is then formed by
quantizing the extracted characteristic values. Since the
Daubechies wavelet transform is used as a wavelet transform, more
accurate characteristic values can be extracted while maintaining
maximum advantage of the wavelet. Furthermore, in a case where the
dimension of the characteristic vector is reduced by quantizing the
extracted characteristic values into binary values, that is, when a
low capacity characteristic vector is formed, a method of measuring
similarity between the weighted registered and inputted
characteristic vectors is properly used to prevent reduction of
acceptance resulting from the formation of the low capacity
characteristic vector. The user authenticity is therefore
determined by the foregoing method.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 is a view showing the constitution of image acquisition
equipment used for performing an iris recognition method according
to the present invention.
FIG. 2 is a flowchart showing procedures for verifying an iris
image according to the present invention.
FIG. 3 is a flowchart showing procedures for multi-dividing the
iris image through Daubechies wavelet transform according to the
present invention.
FIG. 4 shows an example of multi-dividing the iris image through
the Daubechies wavelet transform.
FIG. 5 is a flowchart showing procedures for forming a
characteristic vector of the iris image based on data acquired from
the procedures of multi-dividing the iris image according to the
present invention.
FIG. 6a shows a distribution example of characteristic values of
the extracted iris image.
FIG. 6b shows a quantization function for generating binary
characteristic vector from the distribution example of FIG. 6a.
FIG. 7 is a flowchart showing procedures for determining user
authenticity through a similarity measurement between the
characteristic vectors.
DETAILED DESCRIPTION FOR PREFERRED EMBODIMENT
Hereinafter, a method of recognizing a human iris using a
Daubechies wavelet transform according to the present invention
will be explained in detail with reference to the accompanying
drawings.
FIG. 1 shows the constitution of image acquisition equipment for
use in a method of recognizing a human iris according to the
present invention. Referring to FIG. 1, the constitution of the
iris image acquisition equipment will be explained. The image
acquisition equipment for use in the method of recognizing the
human iris according to the present invention comprises a halogen
lamp 11 for illuminating the iris in order to acquire clear iris
patterns, a CCD camera 13 for photographing an eye 10 of a user
through a lens 12, a frame grabber 14 connected to the CCD camera
12 for acquiring an iris image, and a monitor 15 for showing the
image, which are currently inputted to the camera, to the user so
that acquisition of correct images and positioning convenience of
the user can be obtained when images are acquired.
According to the constitution of the image acquisition equipment,
the CCD camera is used to acquire the image, and iris recognition
is made through a pattern analysis of iridial folds. However, in a
case where the iris image is acquired indoors by using an ordinary
illuminator, it is difficult to extract desired pattern information
since the iris image is generally gloomy. Additional illuminators
should therefore be used so that the information on the iris image
cannot be lost. In such a case, loss of the iris pattern
information and deterioration of recognition capability due to
reflective light should be prevented, and proper illuminators
should be utilized so that a clear iris pattern can be obtained. In
the present invention, the halogen lamp 11 having strong
floodlighting effects is used as a main illuminator so that the
iris pattern can be clearly shown. Further, as shown in FIG. 1, the
loss of the iris image information and eye fatigue of the user can
be avoided by placing the halogen lamp illuminators on the left and
right sides of the eye in order to cause the reflective light from
the lamp to be formed on outer portions of the iris region.
FIG. 2 is a flowchart showing procedures for verifying the iris
image according to the present invention. Referring to FIG. 2, an
eye image is acquired through the image acquisition equipment as
mentioned above in step 200. In step 210, images of the iris
regions are extracted from the acquired eye image-through
pre-processing and transformed into a polar coordinate system, and
the transformed iris pattern is inputted to a module for extracting
the features. In step 220, the Daubechies wavelet transform of the
inputted iris pattern transformed into the polar coordinate system
is performed, and the features of the iris regions are then
extracted. The extracted features have real numbers. In step 230, a
binary characteristic vector is generated by applying K-level
quantization function to the extracted features. In step 240,
similarity between the generated characteristic vector and
previously registered data of the user is measured. Through the
similarity measurement, user authenticity is determined and then
verification results are shown.
In a case where the features of the iris regions are extracted by
performing the Daubechies wavelet transform as described above, the
Daubechies wavelet function having eight, sixteen or more
coefficients can extract more delicate characteristic values than
the Daubechies wavelet function having four coefficients, even
though the former is more complicate than the latter. Although the
Daubechies wavelet function having eight or more coefficients has
been used and tested in the present invention, greater performance
improvement was not obtained from the present invention and
arithmetic quantity and processing time are increased, as compared
with a case where the Daubechies wavelet function having four
coefficients has been used and tested. Thus, the Daubechies wavelet
function having four coefficients has been used for extracting the
characteristic values.
FIG. 3 is a flowchart showing procedures for multi-dividing the
iris image by performing the Daubechies wavelet transform according
to the present invention, and FIG. 4 shows an image divided through
the Daubechies wavelet transform. Referring to FIGS. 3 and 4, in
the present invention, the Daubechies wavelets among various mother
wavelets are used to perform extraction of the iris image
characteristics. As shown in FIG. 4, when "L" and "H" are
respectively used to indicated low frequency and high frequency
components, the term "LL" means a component that has passed through
a low-pass filter (LPF) in all x and y directions whereas a term
"HH" means a component that has passed through a high-pass filter
(HPF) in the x and y directions. Furthermore, subscript numerals
signify image-dividing stages. For example, "LH.sub.2" means that
the image has passed through the low-pass filter in the x direction
and through the high-pass filter in the y direction during 2-stage
wavelet division.
In step 310, the inputted iris image is multi-divided by using the
Daubechies wavelet transform. Since the iris image is considered as
a two-dimensional signal in which one-dimensional signals are
arrayed in the x and y directions, quarterly divided components of
one image should be extracted by passing through the LPF and HPF in
all x and y directions in order to analyze the iris image. That is,
one two-dimensional image signal is wavelet-transformed in vertical
and horizontal directions, and the image is divided into four
regions LL, LH, HL, and HH after the wavelet transform has been
performed once. At this time, through the Daubechies wavelet
transform, the signal is divided into a differential component
thereof that has passed through the high-pass filter, and an
average component that has passed through the low-pass filter
Alternatively, performance of the iris recognition system is
evaluated in view of two factors; a false acceptance rate (FAR) and
a false rejection rate (FRR). Here, the FAR means a probability
that entrance of unregistered persons (imposters) may be accepted
due to false recognition of unregistered persons as registered
ones, and the FRR means a probability that entrance of registered
persons (enrollees) is rejected due to false recognition of the
registered persons as unregistered ones. For reference, when the
method of recognizing the human iris using the Daubechies wavelet
transform according to the present invention is employed, the FAR
has been reduced from 5.5% to 3.07% and the FRR has also been
reduced from 5.0% to 2.25%, as compared with the method of
recognizing the human iris using the conventional Harr wavelet
transform.
In step 320, a region HH including only high frequency components
in the x and y directions is extracted from the divided iris
image.
In step 330, after increasing the iterative number of times of
dividing the iris image, the processing step is completed when the
iterative number is greater than a predetermined number.
Alternatively, if the iterative number is lower than the
predetermined number, the information on the region HH is stored as
information for extracting the iris features in step 340.
Further, in step 350, a region LL comprising only low frequency
components in the x and y directions is extracted from the
multi-divided iris image. Since the extracted region LL
(corresponding to the image reduced in a fourth size as compared
with the previous image) includes major information on the iris
image, it is provided as an image to be newly processed so that the
wavelet transform can be again applied to the relevant region.
Thereafter, the Daubechies wavelet transform is repeatedly
performed from step 310.
On the other hand, in a case where the iris image is transformed
from the Cartesian coordinate system to polar coordinate system, in
order to avoid changes in the iris features according to variations
in the size of the pupil, the region between the inner and outer
boundaries of the iris is divided into 60 segments in the r
direction and 450 segments in the .theta. direction by varying the
angles by 0.8 degrees. Finally, the information on the iris image
is acquired and normalized as 450.times.60 (.theta..times.r) data.
Then, if the acquired iris image is once again wavelet-transformed,
the characteristics of the 225.times.30 region HH.sub.1 of which
size is reduced by half are obtained, namely, the 225.times.30
information is used as a characteristic vector. This information
may be used as it is, but a process of dividing the signals is
repeatedly performed in order to reduce the information size. Since
the region LL includes major information on the iris image, the
characteristic values of further reduced regions such as HH.sub.2,
HH.sub.3 and HH.sub.4 are obtained by successively applying the
wavelet transform to respective relevant regions.
The iterative number, which is provided as a discriminating
criterion for repeatedly performing the wavelet transform, should
be set as a proper value in consideration of loss of the
information and size of the characteristic vector. Therefore, in
the present invention, the region HH.sub.4 obtained by performing
the wavelet transform four times becomes a major characteristic
region, and values thereof are selected as components of the
characteristic vector. At this time, the region HH.sub.4 contains
the information having 84 (=28.times.3) data
FIG. 5 is a flowchart showing procedures for forming the
characteristic vector of the iris image by using the data acquired
from the process of multi-dividing the iris image according to the
present invention. Referring to FIG. 5, the information on the n
characteristic vector extracted from the above process, i.e., the
information on the regions HH.sub.1, HH.sub.2, HH.sub.3, and
HH.sub.4 is inputted in step 510. In step 520, in order to acquire
the characteristic information on the regions HH.sub.1, HH.sub.2
and HH.sub.3 excluding the information on the region HH.sub.4
obtained through the last wavelet transform among the n
characteristic vector, each average value of the regions HH.sub.1,
HH.sub.2 and HH.sub.3 is calculated and assigned one dimension. In
step 530, all the values of the final obtained region HH.sub.4 are
extracted as the characteristic values thereof. After extraction of
the characteristics of the iris image signals has been completed,
the characteristic vector is generated based on these
characteristics. A module for generating the characteristic vector
mainly performs the processes of extracting the characteristic
values in the form of real numbers and then transforming them to
binary codes consisting of 0 and 1.
However, in step 540, the N-1 characteristic values extracted from
step 520 and the M (the size of the final obtained region HH)
characteristic values extracted from step 530 are combined and
(M+N-1)-dimensional characteristic vector is generated. That is,
the total 87 data, which the 84 data of the region HH.sub.4 and the
3 average data of the regions HH.sub.1, HH.sub.2 and HH.sub.3 are
combined, are used as a characteristic vector in the present
invention.
In step 550, the values of the previously obtained characteristic
vector, i.e., respective component values of the characteristic
vector expressed in the form of the real numbers are quantized into
binary values 0 or 1. In step 560, the resultant (M+N-1)-bit
characteristic vector is generated by the quantized values. That
is, according to the present invention, the resultant 87-bit
characteristic vector is generated.
FIG. 6a shows a distribution example of the characteristic values
of the extracted iris image. When the values of the 87-dimensional
characteristic vector are distributed according to respective
dimensions, the distribution roughly takes a shape of FIG. 6a. The
binary vector including all the dimensions is generated by the
following Equation 1. f.sub.n=0 if f(n)<0 f.sub.n=1 if f(n)>0
[Equation 1]
where f(n) is a characteristic value of the n-th dimension and
f.sub.n is a value of the n-th characteristic vector.
When the 87-bit characteristic vector that is obtained by assigning
one bit to the total 87 dimensions are generated in order to use a
low capacity characteristic vector, improvement of the recognition
rate is limited to some extent since loss of the information on the
iris image is increased. Therefore, when generating the
characteristic vector, it is necessary to prevent information loss
while maintaining the minimum capacity of the characteristic
vector.
FIG. 6b shows a quantization function for generating a binary
characteristic vector from the distribution example of the
characteristic values shown in FIG. 6a The extracted
(M+N-1)-dimensional characteristic vector shown in FIG. 6a is
evenly distributed mostly between 1 and -1 in view of its
magnitude. Then, the binary vector is generated by applying the
K-level quantization function shown in FIG. 6a to the
characteristic vector. Since only signs of the characteristic
values are obtained through the process of Equation 1, it is
understood, that information on the magnitude has been discarded.
Thus, in order to accept the magnitude of the characteristic
vector, a 4-level quantization process was utilized in the present
invention.
As described above, in order to efficiently compare the
characteristic vector generated through the 4-level quantization
with the registered characteristic vector, the quantization levels
have the weights expressed in the following Equation 2. f.sub.n=4
if f(n).gtoreq.0.5 (level 4) f.sub.n=1 if 0.5>f(n).gtoreq.0
(level 3) f.sub.n=-1 if 0>f(n)>-0.5 (level 2) f.sub.n=-4 if
f(n)<-0.5 (level 1) [Equation 2]
where f.sub.n means an n-th dimension of the previously registered
characteristic vector f.sub.R of the user or the characteristic
vector f.sub.T of the user generated from the iris image of the eye
image of the user. Explanation of how to use the weights expressed
in Equation 2, is as follows.
In a case where the n-th dimensional characteristic value f(n) is
equal or more than 0.5 (level 4), the value of the i-th dimension
f.sub.Ri or f.sub.Ti is converted and assigned 4 if the value is
"11". In a case where the n-th dimensional characteristic value
f(n) is more than 0 and, less than 0.5 (level 3), the value of the
i-th dimension f.sub.Ri or f.sub.Ti is converted and assigned 1 if
the value is "10". In a case where the n-th dimensional
characteristic value f(n) is more than -0.5 and less than 0 (level
2), the value of the i-th dimension f.sub.Ri or f.sub.Ti, is
converted and assigned -1 if the value is "01". In a case where the
n-th dimensional characteristic value f(n) is equal or less than
-0.5 (level 1), the value of the i-th dimension f.sub.Ri or
f.sub.Ti, is converted and assigned -4 if the value is "00". This
is due to the weights being applied to respective values as
expressed in Equation 2 as it is suitable for the following
verification method of the present invention.
FIG. 7 is a flowchart showing procedures for discriminating the
user authenticity through similarity measurement between the
characteristic vectors. Referring to FIG. 7, in step 710, the
characteristic vector f.sub.T of the user is generated from the
iris image of the eye image of the user. In step 720, the
previously registered characteristic vector f.sub.R of the user is
searched. In step 730, in order to measure the similarity between
the two characteristic vectors, the weights are assigned to the
characteristic vectors f.sub.R and f.sub.T depending on the value
of the binary characteristic vector based on Equation 2.
In step 740, an inner product or scalar product S of the two
characteristic vectors is calculated and the similarity is finally
measured. Among the measures generally used for determining
correlation between the registered characteristic vector f.sub.R
and the characteristic vector f.sub.T of the user, it is the inner
product S of the two characteristic vectors which indicate the most
direct association. That is, after the weights have been assigned
to the respective data of the characteristic vector in step 730,
the inner product S of the two characteristic vectors is used to
measure the similarity between the two vectors.
The following Equation 3 is used for calculating the inner product
of the two characteristic vectors.
.times..times..times..times..times..times..times..times..times..times.
##EQU00001##
where f.sub.R is the characteristic vector of the user that has
been already registered, and f.sub.T is the characteristic vector
of the user that is generated from the iris image of the eye of the
user.
According to the above processes, one effect which can be obtained
by the quantization according to the sign of the characteristic
vector values as in the method in which the binary vector is
generated with respect to the values of the characteristic vector
extracted from the iris image according to respective dimensions
can be maintained. That is, like the Harming distance, the
difference between 0 and 1 can be expressed. In a case where the
two characteristic vectors have the same-signed values with respect
to the each dimension, positive values are added to the inner
product S of the two characteristic vectors. Otherwise, negative
values are added to the inner product S of the two vectors.
Consequently, the inner product S of the two characteristic vectors
increases if the two data belong to an identical person, while the
inner product S of the two characteristic vectors decreases if the
two data does not belong to an identical person.
In step 750, the user authenticity is determined according to the
measured similarity obtained from the, inner product S of the two
characteristic vectors. At this time, the determination of the user
authenticity based on the measured similarity depends on the
following Equation 4. If S>C, then TRUE or else FALSE [Equation
4]
where C is a reference value for verifying the similarity between
the two characteristic vectors.
That is, if the inner product S of the two characteristic vectors
is equal or more than the verification reference value C, the user
is determined as an enrollee. Otherwise, the user is determined as
an imposter.
As described above, the method of recognizing the human iris using
the Daubechies wavelet transform according to the present invention
has an advantage that FAR and FRR can be remarkably reduced as
compared with the method using the conventional Harr wavelet
transform, since the iris features are extracted from the inputted
iris image signals through the Daubechies wavelet transform.
Furthermore, in order to verify the similarity between the
registered and extracted characteristic vectors f.sub.R and
f.sub.T, the inner product S of the two characteristic vectors is
calculated, and the user authenticity is determined based on the
measured similarity obtained by the calculated inner product S of
the two vectors. Therefore, there is provided a method of measuring
the similarity between the characteristic vectors wherein loss of
the information, which may be produced by forming the low capacity
characteristic vectors, can be minimized.
The foregoing is a mere embodiment for embodying the method of
recognizing the human iris using the Daubechies wavelet transform
according to the present invention. The present invention is not
limited to the embodiment described above. A person skilled in the
art can make various modifications and changes to the present
invention without departing from the technical spirit and scope of
the present invention defined by the appended claims.
* * * * *